Two-Staged Algorithm for Eyebrows Segmentation

Eyebrows segmentation technique has been rapidly evolved these years, yet the precision and efficiency of the segmentation algorithm cannot be simultaneously balanced. Instead of treating the segmentation task as an independent problem, we investigate the possibility of balancing segmentation robustness with computationally efficiency through a two stage solution. In this paper, we start from achieving a rough localization of eyebrows based on a facial landmark detection. Next, we generate a dataset based on Helen dataset using this rough localization algorithm, and present an optimized U-Net convolutional network to accomplish the precise segmentation inside of the localized region. Our two-staged model achieved a greater accuracy than traditional facial landmark detection, and a better efficiency than global segmentation task over the whole picture.